You could get clever algorithmically, but I think it would be far simpler to look for identifying traits of specific software packages. eg. if you see links to

You could get clever algorithmically, but I think it would be far simpler to look for identifying traits of specific software packages. eg. if you see links to /wp-content/something, you're looking at a Wordpress site and therefore a blog. Obviously that alone won't be able to classify a lot of sites, but I suspect it will get the bulk of them.